READ ME File For 'Dataset in support of the journal paper 'Predictive Visualisation of Fibre Laser Cutting Topography via Deep Learning with Image Inpainting''

ReadMe Author: Alexander F. Courtier, University of Southampton

This dataset supports the publication:
AUTHORS:Alexander F. Courtier, Matt Praeger, James A. Grant-Jacob, Christophe Codemard, Paul Harrison, Ben Mills, Michalis Zervas
TITLE:Predictive Visualisation of Fibre Laser Cutting Topography via Deep Learning with Image Inpainting
JOURNAL:Journal of Laser Applications
PAPER DOI IF KNOWN:
https://doi.org/10.5258/SOTON/D2489


This dataset contains: A method figure, examples images for experimental and predicted topography sections, as well as examples for a topography inpainting visualisation cGAN. There is also a demonstration of the application of inpainting to increae the size of experimental topographies, as well as using inpainting to predict the appearance of defects between different cutting speeds. There is also a regression plot comparing experimental and predicted labels for a CNN and a collection of histograms and plots comparing the statistical distributions of experimental and inpainted topographic sections, with their respective numerical datasets.

The datasets are contained in 'Datasets:

CONTAINED IN 'Datasets for Figure 3':

Dataset 1.txt  Experimental kerf profies for all speeds in Fig. 3a.
Dataset 2.txt  Experimental roughness profies for all speeds in Fig. 3b.


CONTAINED IN 'Datasets for Figure 6':

Dataset 3.txt  Histogram comparing the pixel intensities of experimental and predicted topography sections of samples cut at 15 m/min and experimental topography samples cut at 16 m/min shown in Fig. 6a.
Dataset 4.txt  Histogram comparing the pixel intensities of experimental and predicted topography sections of samples cut at 20 m/min and experimental topography samples cut at 19 m/min and 21 m/min shown in Fig. 6b.
Dataset 5.txt  Comparison of average kerf profiles of experimental and predicted topography sections of samples cut at 15 m/min and experimental topography samples cut at 16 m/min shown in Fig. 6c.
Dataset 6.txt  Comparison of average kerf profiles of experimental and predicted topography sections of samples cut at 20 m/min and experimental topography samples cut at 19 m/min and 21 m/min shown in Fig. 6d.
Dataset 7.txt  Comparison of average roughness profiles of experimental and predicted topography sections of samples cut at 15 m/min and experimental topography samples cut at 16 m/min shown in Fig. 6e.
Dataset 8.txt  Comparison of average roughness profiles of experimental and predicted topography sections of samples cut at 20 m/min and experimental topography samples cut at 19 m/min and 21 m/min shown in Fig. 6f.


Dataset 9.txt  Mean absolute error of predicted topography as a function of cutting speed for Fig. 7.


CONTAINED IN 'Datasets for Figure 8':

Dataset 10.txt Experimental kerf profile of 15 m/min for Fig. 8a.
Dataset 11.txt Predicted kerf profile of 15 m/min of a CGAN trained on the upper 80% of the available cutting speeds for Fig. 8a.
Dataset 12.txt Predicted kerf profile of 15 m/min of a CGAN trained on the upper 60% of the available cutting speeds for Fig. 8a.
Dataset 13.txt Predicted kerf profile of 15 m/min of a CGAN trained on the upper 40% of the available cutting speeds for Fig. 8a.

Dataset 14.txt Experimental kerf profile of 20 m/min for Fig. 8b.
Dataset 15.txt Predicted kerf profile of 20 m/min of a CGAN trained on the outer 80% of the available cutting speeds for Fig. 8b.
Dataset 16.txt Predicted kerf profile of 20 m/min of a CGAN trained on the outer 60% of the available cutting speeds for Fig. 8b.
Dataset 17.txt Predicted kerf profile of 20 m/min of a CGAN trained on the outer 40% of the available cutting speeds for Fig. 8b.

Dataset 18.txt Experimental roughness profile of 15 m/min for Fig. 8c.
Dataset 19.txt Predicted roughness profile of 15 m/min of a CGAN trained on the upper 80% of the available cutting speeds for Fig. 8c.
Dataset 20.txt Predicted roughness profile of 15 m/min of a CGAN trained on the upper 60% of the available cutting speeds for Fig. 8c.
Dataset 21.txt Predicted roughness profile of 15 m/min of a CGAN trained on the upper 40% of the available cutting speeds for Fig. 8c.

Dataset 22.txt Experimental roughness profile of 20 m/min for Fig. 8d.
Dataset 23.txt Predicted roughness profile of 20 m/min of a CGAN trained on the outer 80% of the available cutting speeds for Fig. 8d.
Dataset 24.txt Predicted roughness profile of 20 m/min of a CGAN trained on the outer 60% of the available cutting speeds for Fig. 8d.
Dataset 25.txt Predicted roughness profile of 20 m/min of a CGAN trained on the outer 40% of the available cutting speeds for Fig. 8d.


CONTAINED IN 'Datasets for Figure 9':

Dataset 26.txt Number of frequencies plotted against the number of inpainting steps for 15 m/min  for Fig. 9a.
Dataset 27.txt Number of frequencies plotted against the number of inpainting steps for 20 m/min  for Fig. 9a.

Dataset 28.txt Mean absolute error for autocorrelation of the center topography against the predicted topography for 15 m/min for Fig. 9b.
Dataset 29.txt Mean absolute error for autocorrelation of the center topography against the predicted topography for 20 m/min for Fig. 9b.

Dataset 30.txt Fourier transform of autocorrelation error for 15 m/min for Fig. 9c.
Dataset 31.txt Fourier transform of autocorrelation error for 20 m/min for Fig. 9d.


CONTAINED IN 'Figure 10':

Dataset 32.txt  Regression plot showing comparing the predicted cutting speeds to the experimental cutting speeds for a regression CNN trained on laser cutting topographies in Fig. 10 b.1.
Dataset 33.txt  Regression plot evaluating the performance of an inpainting cGAN when inpainting between topographies of different cutting speeds using a CNN trained on experimental topographies in Fig. 10 b.2.
Dataset 34.txt  Error plot showing the mean error between predicted cutting speeds and experimental cutting speeds for inpainted topographies produced from topographies of different cutting speeds in Fig. 10 b.3.
Dataset 35.txt  Regression plot evaluating the performance of an inpainting cGAN when inpainting between inpainted topographies of different cutting speeds using a CNN trained on experimental topographies in Fig. 10 b.4.
Pix2Pix repository.txt  Link to github of original Pix2Pix Network by Philip Isola

The figures are contained in 'Figures':

CONTAINED IN 'Figures':

Fig.1.pdf  Data collection and processing method' 
Fig.2.pdf  Examples of topographic sections for stainless steel samples cut at ten different cutting speeds in 2D and 3D.
Fig.3.pdf  Comparison of average kerf profles and average roughness profiles for all cutting speeds
Fig.4.pdf Concept for data augmentation
Fig.5.pdf  Demonstration of the application of image inpainting to expand the length of experimental laser cut topographies.
Fig.6.pdf  Comparison of pixel histograms, average kerf profiles and average roughness profiles of predicted laser cutting topographies cut at 15 m/min and 20 m/min.
Fig.7.pdf Mean absolute error of topography prediciton as a function of cutting speed.
Fig.8.pdf Comparison of prediction capabilities of GANs trained on different portions of the the training dataset.
Fig.9.pdf Analysis of error with regards to repeating patterns in predicted topography
Fig. 10.pdf  Demonstration of the application of image inpainting to predict topographies of the intermediate cutting speeds from samples cut at different speeds.


Date of data collection: 10/2/2021 - 10/3/2021

Information about geographic location of data collection: Data was collected at the University of Southampton, Building 46 (Physics)

Licence:CC-BY

Related projects:
None

Date that the file was created: 14, 12, 2022